Hi, > Am 13.08.2015 um 11:44 schrieb Manfred Selz <[email protected]>: > > Hi Reuti, > > Thanks for commenting. No, not really: > > $ qconf -ssconf > Algorithm default > schedule_interval 0:0:10 > maxujobs 0 > queue_sort_method seqno > job_load_adjustments np_load_avg=0.50 > load_adjustment_decay_time 0:7:30
You can experiment by setting the above to a higher value and longer time. Depending on your job turnaround it might help. -- Reuti > load_formula np_load_avg > schedd_job_info true > ... > > I have seen the page you linked before and played around with the > load_formula settings, but without much success yet. > > My point is that filling up the hosts evenly works really well (mainly based > on host load, as defined by the setting "load_formula np_load_avg") with > "normal" jobs (no parallel environment), but not so well for jobs using our > parallel environment. > > Regards, > Manfred > > > -----Original Message----- > From: Reuti [mailto:[email protected]] > Sent: Donnerstag, 13. August 2015 11:15 > To: Manfred Selz > Cc: [email protected] > Subject: Re: [gridengine users] equal host loading and parallel environment > > Hi, > >> Am 13.08.2015 um 07:53 schrieb Manfred Selz <[email protected]>: >> >> Hi, >> >> I am using SGE 6.2u5 on RHEL5. >> Users in our company run a lot of multi-core simulations via SGE and >> parallel environment like this: >> >> $ qconf -sp local >> pe_name local >> slots 1000 >> user_lists NONE >> xuser_lists NONE >> start_proc_args /bin/true >> stop_proc_args /bin/true >> allocation_rule $pe_slots >> control_slaves FALSE >> job_is_first_task TRUE >> urgency_slots min >> accounting_summary TRUE > > Did you set up any particular scheduling parameters: > > https://blogs.oracle.com/sgrell/entry/grid_engine_scheduler_hacks_least > > -- Reuti > > >> >> The jobs go into this queue: >> >> $ qconf -sq mmtoken.q >> qname mmtoken.q >> hostlist @nb-lin @nb-lin-aux @nb-lin-mmtoken >> seq_no 5,[@nb-lin-mmtoken=1],[@nb-lin-aux=10] >> load_thresholds np_load_avg=0.75,localdisk_free=300G,mem_free=3G, \ >> mmtoken=20 >> suspend_thresholds NONE >> nsuspend 1 >> suspend_interval 00:05:00 >> priority 0 >> min_cpu_interval 00:05:00 >> processors UNDEFINED >> qtype BATCH INTERACTIVE >> ckpt_list NONE >> pe_list local make >> ... >> >> According to documentation, I’d expect that the allocation rule “$pe_slots” >> helps to fill the hosts more or less equally, with priorities defined by the >> different seq_no settings in the queue. >> However, I frequently see that hosts are filled unevenly, and new jobs with >> multiple slots are often added to hosts already running other jobs instead >> of being assigned to hosts with a lower load. >> Does anybody happen to know how this could be influenced? >> >> Regards, >> Manfred >> >> _____________________________________________________________ >> Manfred Selz >> Senior CAD Engineer >> Direct Dial: +49 (0)7021 805-562 >> [email protected]| www.diasemi.com Dialog Semiconductor GmbH, >> Neue Strasse 95, 73230 Kirchheim/Teck-Nabern, Germany >> _____________________________________________________________ >> >> >> >> >> Dialog Semiconductor GmbH >> Neue Str. 95 >> D-73230 Kirchheim >> Managing Directors: Dr. Jalal Bagherli, Jean-Michel Richard Chairman >> of the Supervisory Board: Rich Beyer Commercial register: Amtsgericht >> Stuttgart: HRB 231181 UST-ID-Nr. DE 811121668 >> >> Legal Disclaimer: This e-mail communication (and any attachment/s) is >> confidential and contains proprietary information, some or all of which may >> be legally privileged. It is intended solely for the use of the individual >> or entity to which it is addressed. Access to this email by anyone else is >> unauthorized. If you are not the intended recipient, any disclosure, >> copying, distribution or any action taken or omitted to be taken in reliance >> on it, is prohibited and may be unlawful. >> >> >> Please consider the environment before printing this e-mail >> >> _______________________________________________ >> users mailing list >> [email protected] >> https://gridengine.org/mailman/listinfo/users > > ________________________________ > > Dialog Semiconductor GmbH > Neue Str. 95 > D-73230 Kirchheim > Managing Directors: Dr. Jalal Bagherli, Jean-Michel Richard > Chairman of the Supervisory Board: Rich Beyer > Commercial register: Amtsgericht Stuttgart: HRB 231181 > UST-ID-Nr. DE 811121668 > > Legal Disclaimer: This e-mail communication (and any attachment/s) is > confidential and contains proprietary information, some or all of which may > be legally privileged. It is intended solely for the use of the individual or > entity to which it is addressed. Access to this email by anyone else is > unauthorized. If you are not the intended recipient, any disclosure, copying, > distribution or any action taken or omitted to be taken in reliance on it, is > prohibited and may be unlawful. > > Please consider the environment before printing this e-mail > > > _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
